Robert Gordon has argued in his recent NBER paper “Is U.S. Economic Growth Over? Faltering Innovation Confronts the Six Headwinds” that growth rates have slowed and we are reverting to very low historical growth rates and indeed a period of economic stagnation. However, what I find intriguing is that an examination of some long-term data for Britain provides conflicting results that quite frankly puzzle me.
Greg Clark has constructed a long-term series for average annual real earnings in Britain stretching from 1209 to 2010. (See G. Clark, “What Were the British Earnings and Prices Then? (New Series)” Measuring Worth, 2011). As Figure 1 shows, there was a period of growth in average annual real earnings (2010 pounds) from about the early 1300s to the mid 1400s, followed by a long decline with growth not resuming until the advent of the industrial revolution after which real earnings have increased dramatically.
Figure 1
If one calculates and plots the the annual growth rates and then does a LOWESS smooth (0.8 bandwidth) on the growth rates, one obtains Figure 2a. Figure 2b plots only the LOWESS smooth and provides a much better picture of what the course of growth rates (in percent) for average annual real earnings have been. What is interesting about Figure 2b is that growth rates (in percent) for real average earnings in Britain have increased steadily since about the mid 1600s. Indeed, there appears to be no decline during the course of the twentieth century.
Figure 2a
Figure 2b
This is in contrast to the result you get when using growth rates for real per capita GDP. Lawrence Officer and Sam Williamson provide a real per capita GDP series for the UK for the period 1830 to 2010 (L.H. Officer and S.H. Williamson, “What Was the U.K. GDP Then?” Measuring Worth 2011). When the growth rates for real per capita GDP are calculated and a LOWESS smooth applied, you get the result that Robert Gordon has made the argument for. Growth rates for real per capita GDP in the UK rise from about the mid-19th century and peak after 1945 at about 2 percent and then begin to decline. Figure 3 takes the LOWESS smoothes for the growth rates for both average annual real earnings (1245-2010) and real per capita GDP (1831-2010) and plots them together.
Figure 3
This is a bit of a puzzle. The earnings growth rates show a pretty steady upward trend even into the present while real per capita GDP growth rates shows a decline setting in after the middle of the 20th century. Based on the earnings profile, it has been onward and upwards for the last 400 years and there does not yet seem to be an end in sight. Does earnings growth perhaps lag per capita GDP growth and decline is just around the corner? If so, why? Why is earnings growth so steady relative to per capita GDP growth? Is there a labor economist out there who might shed some light on this?
Livio, in answer to your question "why is earnings growth so steady relative to GDP growth?" Earnings are defined as earnings per full-time worker. GDP is GDP per person, including employed, unemployed and not in the labour force. In a recession, a lot of people lose their jobs. That affects real per capita GDP, but has much less impact on earnings per worker, so GDP growth is much bumpier than earnings growth. (Also - one of the big trends in the labour market is towards more part-time and casual work. That's not picked up in those FT earnings numbers.)
You also asked "Does earnings growth perhaps lag per capita GDP growth and decline is just around the corner?" This gets back to issues we've talked a lot about here - population aging, demographics, and turning Japanese. In 15 years time, when the vast bulk of the baby boomers are over 65, the percentage of the population who are working and earning will almost certainly be a smaller than it is today. Unless earning per full-time worker increase substantially, per capita GDP will fall. (This is one reason some people obsess about productivity and others are so strongly pro-immigration - they hope that this will allow them to avoid the inevitable: population aging + current patterns of labour force participation = decline in per capita GDP).
Posted by: Frances Woolley | February 07, 2013 at 01:25 AM
absolute income goes down from 3500 to 1800 (per thumb) from 1450 to 1650 in Figure 1 (that would be MINUS 0.35%), but growth rate in Figure 2b is PLUS 0.5% ? How come ?
Posted by: genauer | February 07, 2013 at 04:15 AM
Frances:
Well, the good news then is that productivity per worker has been rising over the long term at least as measured by earnings.
Genauer:
Good question. I do not expect exact correspondence between the two figures. Figure 1 is average annual earnings. Figure 2b is a LOWESS smooth of growth rates for those earnings. The smooth is done with a 0.8 bandwidth which means 80% of the observations available around the point being estimated are used in the calculation and there are declining weights as the observations move away from the point. However, there are also declining rates of growth in 2b over approximately the same period.
Posted by: Livio Di Matteo | February 07, 2013 at 07:44 AM
Livio - true, that is good news - and that's an interesting question - is the stagnation that Gordon predicts an end of growth in GDP/capita, or an end of growth in GDP/worker - two entirely different things, with entirely different causes/consequences?
Posted by: Frances Woolley | February 07, 2013 at 08:29 AM
France, I believe Gordon was looking at GDP and per capita GDP growth.
Posted by: Livio Di Matteo | February 07, 2013 at 08:39 AM
I think that :
a) the stated plot descriptions are pretty lengthy & specific "Average Annual Real earnings", and
b) the analysis does NOT pass a very elementary smell test.
c) that this does reflect on the quality of the institutions involved in this
Posted by: genauer | February 07, 2013 at 09:39 AM
sorry,
I somehow got the impression that the plots were published data.
A LOWESS filter with parameters, which give over 200 years (year 1450 - 1650) a significantly positive number of 0.5%/anno growth, while the real numbers decay by something like a factor 1.5 - 2 over that period, is not that useful, and then isn't that either for the last 200 years.
But blogs, mimeos and working papers have the purpose, to incite such comments early on.
I was a little grumpy, that I couldnt get my hands immediately on: Goldsmith, A Study of Saving in the United States, Vol. 3 (Princeton, 1956), 20-21
triggered by the recent post of Nick Rowe about "inflation finally falling?"
Posted by: genauer | February 07, 2013 at 11:03 AM
No problem Genauer. I see one of the roles of a blog as a way of getting some pretty preliminary analysis with data out there to see what people might think. I suppose I could experiment with different bandwidths to allow for more local variation in the plots. The 0.8 bandwidth is a maximum smoothing parameter.
Posted by: Livio Di Matteo | February 07, 2013 at 12:11 PM
Without detracting from the comments about aging etc., I wonder if this could also reflect the growth of Britain's financial sector. I am not quite clear on what goes into GDP. If Britain produces less goods domestically but earns money through investing in goods produced elsewhere, how would that affect GDP? I presume that the earnings of financial sector workers would still go into the "earnings per worker" calculation.
Posted by: Paul Friesen | February 08, 2013 at 12:27 PM
Although I've a fair bit of work on the impact of smoothing filters on estimates of recent trends of productivity growth, I'm not familiar with LOWESS. I know other filters (like Hodrick-Prescott or bandpass or state-space models) tend to behave quite differently near the end of samples than they do elsewhere in the sample; in particular, they become more erratic and exhibit phase-shift. All this implies that you should be very skeptical about estimates of recent trends produced by these methods unless you have some exact measures of the size of these distortions.
Anyone know whether LOWESS has similar problems? Is Livio just smoothing over time?
Posted by: Simon van Norden | February 08, 2013 at 02:23 PM
Livio, are the two smoothed series that you show in figure 3 both based on the same frequency of data (annual?) and the same smoothing parameter? I'm wondering whether the two lines would agree if I just increased the degree of smoothing on the shorter series.
Posted by: Simon van Norden | February 08, 2013 at 02:28 PM
Hi Simon: Both series are annual and 0.8 was the bandwidth for both. There are border/edge effects at the start and end of the series in the estimation which might be more important for the GDP series as it is a much shorter series but my econometrics is not deep enough to know how much of an effect there might be.
Posted by: Livio Di Matteo | February 08, 2013 at 05:41 PM